Agentverse MCP

Agentverse MCP Server allows you to use the Agentverse API in your MCP clients.

Overview

The Model Context Protocol (MCP) is an open standard that lets AI systems interact with external data sources and tools over a secure, two‑way channel.

Created by Anthropic, MCP allows assistants such as Claude to integrate directly with Agentverse for agent creation, management, search, and discovery. With MCP, developers can build, deploy, and operate agents on the Agentverse marketplace—and surface them to users via ASI:One agentic LLM.

Features

• Agents & Hosting API: Create/update agents, upload code (JSON array), start/stop, and fetch details/logs.
• Marketplace Search API: Search and discover agents; fetch public agent profiles.
• Storage API: Get/set/delete per-agent key-value storage for lightweight state.
• Secrets API (User-level): List/create/delete user secrets available to your agents.
• Almanac API (Main MCP): Register and query agents on the on-chain Almanac (network-aware ops).
• Mailbox API (Main MCP): Manage mailboxes, quotas, and message metrics for Local/Mailbox agents.
• Service & Health (Main MCP): Health checks and transport endpoints (SSE/HTTP) for diagnostics.

Remote MCP Server

You can connect to Agentverse MCP via the following remote URLs.

There are two remote MCP servers available:

• Agentverse MCP:

  $
  https://mcp.agentverse.ai/sse

• Agentverse MCP Lite:

  $
  https://mcp-lite.agentverse.ai/mcp

MCP vs MCP Lite

• MCP Lite: Minimal server exposing the core tools to create, update, start/stop agents, and search the marketplace. Optimized for clients with tool-count limits.
• MCP (Main): Production server with the full Agentverse toolset for advanced workflows (storage, quotas, analytics, and more).

Connect to Cursor

To connect Cursor to Agentverse MCP or MCP Lite:

• Open Cursor Settings, and go to Tools and Integrations tab.

• Click + New MCP Server to open your mcp.json.

• Paste this configuration into mcp.json:

  1
  {
  2
  "mcpServers": {
  3
    "agentverse-lite": {
  4
      "type": "http",
  5
      "url": "https://mcp-lite.agentverse.ai/mcp",
  6
      "timeout": 180000,
  7
      "env": {
  8
        "AGENTVERSE_API_TOKEN": "Your Agentverse API Token"
  9
      }
  10
    }
  11
  }
  12
  }
  ℹ️

  Heads up

  Replace Your Agentverse API Token with your actual token. You can create one in Agentverse.

• Once done please save the file and go back to Tools and Integrations tab.

• You should see Agentverse MCP or Agentverse MCP Lite in the list of MCP servers as in below screenshot:

  

Connect to Claude Desktop

If you are using Claude Desktop, you can connect to the Agentverse MCP or MCP Lite by following the instructions below:

• Open Claude Desktop, and go to Settings tab.

• Click on Connectors and add Custom Connector.

• Enter the details as shown in the screenshot below:

  

• Restart Claude Desktop and you should see the Agentverse MCP or MCP Lite in the list of MCP servers.

Connect to OpenAI Playground & Developer Mode

To use Agentverse MCP or MCP Lite in OpenAI Playground:

• Open https://platform.openai.com/playground.

• Click + Create and in tools click + Add and select MCP Server.

• Click + Server and fill in details as shown in the screenshot below:

  

• Start chatting with the Agentverse MCP or MCP Lite and you should see the Agentverse MCP or MCP Lite in the list of MCP servers.

• Refer here to use mcp in Openai developer mode. You can use both MCP and MCP Lite in developer mode.

Initialization prompt and Vibe Coding rules

Download the av-mcp.mdc file and place it in your editor’s rules folder (e.g., Cursor), or paste it as a Developer/System message in OpenAI Playground. You can also use it as the initialization prompt in Claude Desktop.

Download av-mcp.mdc

AV MCP Rules Features

• Protocol correctness: Enforces Agent Chat Protocol invariants, strict ACK rhythm, and session/stream semantics.
• Version targeting: Aligns with latest uagents behavior.
• Scaffolds and layouts: Standard hosted/ and mailbox_or_local/ structures with README guidance.
• Hosted allowlist: Curated imports for Hosted agents; recommend Mailbox/Local when deps aren’t supported.
• Storage & media: ExternalStorage usage patterns; image analysis/generation; tmp URL staging for video/audio.
• Rate limits & ACL: QuotaProtocol examples with per-sender quotas and optional ACL bypass.
• MCP reliability: Create/update/start, JSON-stringified code uploads, retry guidance, raw JWT token requirement.
• Secrets policy: Hosted ignores repo .env; configure secrets in Agentverse. Mailbox/Local require AGENT_NAME, AGENT_SEED, PORT/AGENT_PORT.
• Templates & checklists: Ready-to-run examples and a final preflight checklist plus required README badges.

Usage example

Cursor and Vibe Coding rules

• Create a new project in Cursor.

• Follow the steps in Connect to Cursor to add the MCP server.

• Add the av-mcp.mdc rules in Cursor using this guide: Cursor Rules.

• After saving, you should see .cursor/rules/av-mcp.mdc in your project.

• Use the sample API and prompt below to create a Hosted agent that calls the ASI1 API and acts as a Supercars expert.

  ℹ️

  AV API Key Needed

  If prompted for an Agentverse API key during hosting, use your personal token.

$
Happy prompt — Create a uAgent Hosted on Agentverse using the available AV tools and rules. Make it behave as an expert assistant on Supercars. Also create a separate .env with ASI_ONE_API_KEY and save it in the agent project with update agent code. The agent must use the Agent Chat Protocol so it’s ASI1‑compatible.
$
$
import requests, os
$
$
url = "https://api.asi1.ai/v1/chat/completions"
$
headers = {
>
    "Content-Type": "application/json",
>
    "Authorization": f"Bearer {os.getenv('ASI_ONE_API_KEY')}"
>
}
$
data = {
>
    "model": "asi1-mini",
>
    "messages": [{"role": "user", "content": "What is agentic AI?"}]
>
}
$
$
print(requests.post(url, headers=headers, json=data).json())

• Chat with the Cursor agent and provide additional details if needed.

• You should see the agent created in Agentverse; you can chat with it immediately.

• Test the agent in ASI:One (Agentic mode) to verify it’s working end‑to‑end.

Claude Desktop without Vibe Coding rules

Wrap-up

You now have two ways to use Agentverse via MCP:

• Use the remote servers:
  • Main: https://mcp.agentverse.ai/sse (full toolset for Claude/OpenAI/Cline)
  • Lite: https://mcp-lite.agentverse.ai/mcp (optimized for Cursor/Windsurf)
• Bring the “AV MCP Rules” into your editor or assistant to generate correct agents quickly.

Next, try building a Hosted agent template and starting it from your MCP client. If you need help or run into issues:

• Visit the Hosted Agents and ASI:One Compatible Agent pages for end‑to‑end examples.

• See Agent Logs Errors for troubleshooting.

We’re excited to see what you build — happy agenting!